基于最小方差的K-means用户聚类推荐算法  

K-means User Clustering Recommendation Algorithm Based on Minimum Variance

作　　者：杨大鑫 王荣波[1] 黄孝喜[1] 谌志群[1] 

机构地区：[1]杭州电子科技大学计算机学院,浙江杭州310018

出　　处：《计算机技术与发展》2018年第1期104-107,共4页Computer Technology and Development

基　　金：国家自然科学基金青年项目(61202281);教育部人文社会科学研究青年基金项目(10YJCZH052)

摘　　要：协同过滤推荐算法是一种传统的推荐技术,具有简单高效的特点,在实际中有广泛的应用,获得了大量研究者的青睐。虽然传统的协同过滤推荐算法在一定程度上缓解了用户当前所面临的信息超载问题,但其在处理大数据时存在的数据稀疏性和扩展性等问题却日益突出。于是,提出了一种基于最小方差的K-means用户聚类推荐算法。在缓解数据稀疏性方面,利用Weighted Slope One算法对初始用户—项目评分矩阵进行有效填充,降低了数据稀疏性;在提高算法扩展性方面,采用基于最小方差的K-means算法对用户评分数据进行聚类,将相似的用户聚到一起,减小目标用户的最近邻搜索空间,提高了算法扩展性。通过在Movie Lens数据集上的对比实验,结果表明,相比于传统的协同过滤推荐算法,改进算法具有更高的推荐准确度。Collaborative filtering recommendation algorithm is a kind of traditional recommendation technology which is so simple and ef- ficient with a wide range of applications that it has been favorite by a large number of researchers. Although the traditional collaborative filtering recommendation algorithm has alleviated the information overload faced by users to a certain extent,the data sparsity and expan- sibility in dealing with large data is becoming more and more prominent. For this, a K -means user clustering recommendation algorithm based on minimum variance is proposed. The Weighted Slope One algorithm is used to fill the initial user-item scoring matrix effectively, and the data sparsity is reduced. Then K -means algorithm based on minimum variance is adopted to carry out the user rating data cluste- ring,with similar users clustered together to reduce the target user＇ s nearest neighbor search space and improve its expansibility. The con- trast experiments on MovieLens datasets show that the proposed algorithm has higher recommendation accuracy than the conventional col- laborative filtering recommendation algorithm.

关 键 词：信息过载 协同过滤算法 Weighted SLOPE One 最小方差 K—means聚类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]